Recommendation for Nvidia GPU

As a speech recognition researcher group, we are in a need of GPU for Deep Neural Network training. We receive 1 hour of labelled data for each week. Currently we have 26 hours of data. Which GPU would be sufficient for training processes considering capability and money?
Share experiences if you have also researched about DNN.

Speech recognition is somewhat outside my area, but I’d suggest at least getting a RTX card. They have tensor cores that make them much better-suited to deep learning than the GTX-series cards. So, that means a RTX 2060, at the very minimum.

This might help you gauge the relative performance of different models:

You’ll want to pay attention to the first number, in the “Tensor compute” column, for each model.

Make sure your power supply is capacious enough (typically, the product details of a given GPU card will stipulate recommended PSU rating, in Watts - you can find this labeled on the side of your PSU, if you open up the case). Also, check that your case is large enough - most card are 2 slots wide and their length should also be specified in the product details. Don’t forget to connect the auxiliary power, which might require you to get an extra 6-pin, 8-pin (or combined) cable.